Abstract
Simulating protein structure flexibility using classical methods is computationally demanding, especially for large proteins. To address this challenge, we have been developing the CABS-flex method, which enables fast simulations of protein structural flexibility by combining a coarse-grained simulation approach with all-atom detail. Previously available as the CABS-flex 2.0 web server, the method has now undergone a major upgrade with the release of CABS-flex 3.0. Key improvements include the introduction of intuitive flexibility modes that simplify the control of distance restraints and allow users to reflect known or expected dynamic regions; improved all-atom reconstruction for higher-quality model generation; a new feature for de novo peptide structure prediction, supporting both linear and cyclic peptides along with their conformational flexibility; and new tools for result analysis and visualization, facilitating deeper insights into structural flexibility. Additionally, AlphaFold pLDDT-derived restraints can be used as optional input for guiding simulations. The method accepts input as either a PDB/mmCIF structure or a sequence (for peptide modeling). Advanced options allow users to incorporate experimental or computational restraints. The CABS-flex 3.0 web server is available at https://lcbio.pl/cabsflex3. This website is free and open to all users, with no login requirement.
Graphical Abstract
Graphical Abstract.
Introduction
Protein structural flexibility is essential for their biological functions, yet studying protein dynamics experimentally is often challenging or even impossible. Consequently, computational simulations have become indispensable. However, classical approaches, such as all-atom molecular dynamics (MD), are constrained by their high computational cost, making the simulation of structural flexibility for most biologically relevant proteins particularly demanding. Performing such simulations typically requires specialized hardware, such as supercomputers, or substantial computational resources.
To address these challenges, we introduced CABS-flex approximately a decade ago as a method designed for rapid simulations of protein flexibility. CABS-flex offers a computational speed advantage of three to four orders of magnitude over all-atom MD while maintaining competitive accuracy in predicting structural flexibility [1]. Built upon the well-established CABS coarse-grained model, this approach effectively captures key dynamic properties observed in experimental and MD-derived datasets [1–3]. Its integration with all-atom reconstruction enables a seamless transition from coarse-grained flexibility simulations to detailed atomic-level representations. Over the years, we have continuously refined the method, introducing two iterations of the CABS-flex web server [4, 5], a standalone application with enhanced control and scalability [6], and, more recently, extensions that support peptide modeling and refined structural analysis [7]. Today, CABS-flex has gained widespread adoption, particularly through its popular web server 2.0, making flexibility simulations more accessible to the scientific community. Furthermore, it has become an integral component of various computational protocols, enabling researchers to explore the interplay between protein structure, flexibility, and function across diverse fields, as summarized in a recent review [8].
Previous versions of CABS-flex, although widely adopted, had several limitations. These included limited options for customizing flexibility restraints, no dedicated support for peptide modeling, and an all-atom reconstruction method that could lead to suboptimal side-chain conformations and local backbone geometry. Moreover, earlier versions did not incorporate structural confidence metrics such as AlphaFold's pLDDT scores. CABS-flex 3.0 addresses these issues by introducing new flexibility modes with adjustable restraint schemes, a protocol for modeling both linear and cyclic peptides, and an advanced deep learning-based all-atom reconstruction method (cg2all) that substantially improves structural detail [9]. The integration of pLDDT-derived restraints further enhances the biological relevance of flexibility predictions. Together, these improvements significantly expand the scope, accuracy, and usability of CABS-flex for modeling the dynamic behavior of proteins and peptides. These enhancements build on our recent research and are introduced in the “Materials and methods” section.
Materials and methods
CABS-flex combines a fast coarse-grained modeling engine with all-atom modeling, providing an efficient approach for simulating protein flexibility. Our recent review [8] provides a comprehensive overview of the CABS-flex methodology, its applications in structure-function studies, and its role in investigating the relationship between protein structure, dynamics, and function. The current version (3.0) of the CABS-flex web server builds upon the capabilities of its predecessor, CABS-flex 2.0 [5], as well as the CABS-flex standalone application [6], introducing significant improvements in structural flexibility modeling and a new peptide modeling mode. These advancements in CABS-flex 3.0 are based on recent studies, which we briefly reference below as the foundation for its key developments.
In CABS-flex 3.0, we implemented a new approach to distance restraints, as described in a recent study [3]. This work introduces a classification system for different flexibility categories, assigning distance restraints based on secondary structure and AlphaFold-derived pLDDT scores. This methodology enhances the predictive accuracy of CABS-flex while maintaining computational efficiency. We validated this approach by comparing fluctuation profiles obtained from CABS simulations with data from the ATLAS database [10], which contains MD simulations of nearly 1400 proteins. By leveraging pLDDT scores, which indicate the confidence of AlphaFold’s structure predictions, CABS-flex achieves improved accuracy in structural flexibility modeling while preserving its speed advantage over traditional all-atom MD simulations [3]. Nevertheless, when pLDDT scores are unavailable or unreliable, users can choose from the established Flexible and Rigid modes, which do not depend on AlphaFold predictions. These modes have been extensively benchmarked in previous studies [1–3, 8], demonstrating reliable performance across a wide range of protein systems.
Another major addition in CABS-flex 3.0 is the peptide modeling protocol, which was developed using the CABS-flex standalone application [6] and is now incorporated into the web server. This protocol has been benchmarked against state-of-the-art tools such as AlphaFold, PEP-FOLD, APPTEST, and ESMFold [7]. While AlphaFold generally provided the highest accuracy, CABS-flex proved competitive, particularly for short linear peptides. The proposed approach also enables the modeling of cyclic peptides and allows the incorporation of disulfide bridge information. Apart from structural prediction, the protocol also provides insights into peptide flexibility.
A further significant improvement in CABS-flex 3.0 is the integration of a deep-learning-based cg2all method for all-atom reconstruction [9]. In our practice, cg2all provides a substantial enhancement over the previous reconstruction approach based on Modeller [11], particularly in terms of local structural quality, and we consider it the best currently available method. This improvement likely results from cg2all's ability to learn complex side-chain and backbone reconstruction patterns beyond what is possible with traditional optimization methods. The cg2all approach employs an SE(3) transformer architecture and a rigid-body block representation inspired by AlphaFold2. Its integration into CABS-flex 3.0 significantly enhances the accuracy and reliability of reconstructed structures, improving both visualization and downstream analyses of protein flexibility.
Web server input options
The CABS-flex 3.0 web server is designed to operate in two distinct modes: flexibility modeling and peptide modeling, see pipeline in Fig. 1. For ease of use, each mode is accessible through its own dedicated tab on the main page, where users can enter their input data.
Figure 1.
CABS-flex 3.0 pipeline. The image shows some of the input options. The primary output, which is then analyzed using the web interface, includes the simulation trajectory and 10 models..
Flexibility modeling
Basic input options
The only required input for flexibility modeling is the protein structure, provided either as a PDB code or an uploaded file in PDB/mmCIF format. Due to computational limitations, input structures must contain no more than 2000 residues and may include up to 10 chains. Users may optionally select specific chains using the “Chain(s)” field. To help manage submitted jobs, additional optional fields include a “Project name” (displayed in the queue list unless the “Do not show my job on the results page” option is selected) and an “Email address” for receiving job completion notifications.
Flexibility modes
CABS-flex 3.0 introduces four flexibility modes—Flexible, Rigid, Rigid-pLDDT, and Unleashed—that define how distance restraints are applied across the structure and allow users to control the extent of conformational sampling. A representative example of their effect on backbone fluctuations is shown in Fig. 2. Each mode may be appropriate for different modeling contexts, with names that broadly reflect the expected levels of structural flexibility. All modes have been benchmarked against MD and experimental data, as described below:
Figure 2.
Example of the effect of distance restraint mode versus backbone fluctuations in the simulation. The figure presents, from the top left: the starting protein structure (PDB ID: 2f60 chain K) colored according to pLDDT, followed by the output structures obtained using restraints modes: Rigid, Rigid-pLDDT, Flexible, and Unleashed. Further details on the presented restraints were recently discussed in [3].
Flexible mode applies distance restraints only to residues forming secondary structure elements, allowing greater flexibility in loops and unstructured regions. Previously referred to as “SS2” in earlier versions of CABS-flex 2.0 [5] and standalone application [6], it provides a balanced representation of protein dynamics. This mode has been validated against crystallographic B-factors and atomistic MD simulations using various force fields [1], NMR ensembles [2], as well as in numerous structure-flexibility-function studies supported by experimental evidence, including Cryo-EM-derived conformational variability and functional analyses, related to aggregation propensity and S-nitrosylation sensitivity [8]. Importantly, the ability of the CABS model to predict loop structures ab initio has been demonstrated across various protein systems [12, 13], including in the context of peptide interactions [14], highlighting its suitability for modeling local structural variability.
Rigid mode imposes uniform restraints on all residues, minimizing fluctuations throughout the structure. Equivalent to the ‘All’ mode described previously [5, 6], it effectively preserves native-like conformational constraints. Benchmarking with CHARMM36m-based ATLAS MD simulations showed high correlation with MD-derived fluctuation profiles, particularly in structured, globular proteins [3]. It is worth noting that the selection criteria of the ATLAS dataset favor well-folded, high-resolution monomeric proteins, which may bias the dataset toward more structurally stable and less flexible systems. However, some degree of intrinsic flexibility is still likely present in the selected structures.
Rigid-pLDDT mode improves flexibility predictions by integrating structural confidence into the simulation process. In this mode, restraint strength is modulated according to per-residue pLDDT scores and secondary structure classification, as introduced in our recent work [3]. It is applicable when pLDDT scores are available, as in AlphaFold-predicted structures. Users may supply pLDDT values either through the B-factor field in the PDB file or via an external .json or .tsv file. Based on general observations, the overall level of flexibility generated by this mode tends to be closer to Rigid than Flexible, which is reflected in the naming convention, and the mode has been validated using ATLAS MD simulation data [3].
Unleashed mode applies no restraints, allowing for fully unrestricted conformational sampling. The resulting ensemble is governed solely by the intrinsic properties of the CABS coarse-grained force field. This mode is designed as an advanced option, recommended primarily for exploratory simulations where it is important to observe the behavior of a system in the complete absence of restraints. While this mode typically shows lower agreement with MD-derived fluctuation profiles and may produce exaggerated motions, it is valuable for modeling folding/unfolding processes or large-scale transitions in disordered or flexible systems. In such cases, meaningful results may require careful tuning of additional simulation parameters, such as temperature and number of cycles. The applicability of CABS-based simulations to disordered and unfolded protein systems has been reviewed in our previous work [15], which discusses several case studies involving disordered binding partners and unstructured regions.
Mode selection guidance
By default, Flexible mode is enabled in CABS-flex 3.0 due to its versatility and proven accuracy. For stable, well-folded proteins, the Rigid mode may be most appropriate. When pLDDT scores are available, we recommend using Rigid-pLDDT, which incorporates structural confidence into simulations and improves predictive accuracy. Conversely, for systems expected to undergo large-scale motions or unfolding, the Unleashed mode can be used, with caution and additional parameter tuning. When in doubt, we recommend running simulations with multiple flexibility modes, particularly Rigid, Flexible, and Rigid-pLDDT, and comparing the resulting fluctuation profiles with available structural or functional knowledge about the protein. This comparative approach can help identify the most appropriate mode for a given system.
Advanced flexibility control
CABS-flex 3.0 also offers a residue-level flexibility editor that allows users to manually assign flexibility categories prior to simulation. This option, available through a checkbox before job submission, uses the same underlying scheme as the Rigid-pLDDT mode [3] and enables expert-level customization when specific flexibility patterns are known.
Advanced input options
Advanced options allow users to modify default settings based on their needs and available information about the modeled system. Below we outline the main categories of advanced input options available in CABS-flex 3.0. Detailed explanations, usage tips, and practical recommendations are available in the “How To” page (https://lcbio.pl/cabsflex3/howto) on the CABS-flex website. The influence of key parameters, such as restraint schemes and temperature, on simulation outcomes has also been systematically analyzed in our recent study [3].
Flexibility Restraint Settings—a set of restraints is automatically generated based on the selected Flexibility Mode. However, additional parameters in this section allow users to modify the rule set for generating restraints, enabling fine-tuned control over how flexibility is applied in the simulation.
Custom Restraints—users can add restraints beyond the automatically generated ones, incorporating experimental data. CABS-flex 3.0 supports restraints for both Cα and side chains, which can be entered manually or uploaded as a text file.
Simulation Settings—this section provides control over general simulation parameters, such as trajectory length and temperature. Temperature is one of the key factors affecting protein flexibility.
Peptide modeling
The only required input is the peptide sequence. If secondary structure information is not provided, it will be predicted using the NetSurfP-3.0 method [16]. Alternatively, users can provide the secondary structure in HEC format (H for helix, E for beta-strand, C for coil) alongside the sequence, separated by a colon (e.g. ALALA:CHHHH). Additional options include “Add disulfide bonds,” which allows users to select cysteines from the sequence, and “Model cyclic backbone.” Both options act at the coarse-grained simulation level and during all-atom reconstruction with Modeller, ensuring the correct closure of disulfide and/or cyclic backbone bonds in the final structure.
Similar to the flexibility mode, users can specify a “Project name,” choose “Do not show my job on the results page,” provide an “Email address,” and set a “Seed for random number generation.” As in the flexibility modeling mode, CABS-flex generates 10 final models, representing alternative predictions, which can also illustrate the flexibility of the studied peptides.
This modeling protocol was previously benchmarked and evaluated in detail in a recent study [7], which included comparisons with other state-of-the-art tools such as AlphaFold and PEP-FOLD. The benchmark included linear and cyclic peptides, with varying lengths and different types of cyclization. While CABS-flex 3.0 performs best for short, linear, and flexible peptides, it also demonstrates reliable performance for other peptide types, such as cyclic peptides and longer peptides with more complex structures. As of now, the protocol supports only canonical residues and does not include post-translational modifications. These limitations, along with possible workarounds, are discussed in detail in Section 3.7 (Merging with High-accuracy All-atom Reconstruction) of our recent review [8]. Example predictions from this protocol are shown in Fig. 3.
Figure 3.
Example structure predictions of, from the top: linear, cyclic disulfide, and cyclic backbone peptides. On the left, best predicted structure versus experimental structure (lowest RMSD) and on the right, 10 final models superimposed on each other. The method’s performance has been described in [7].
Web server output data
Upon job completion, users gain access to four analysis tabs: Project information, Models, Contact Maps, and Fluctuation Plot, which integrate advanced visualization features (see Fig. 4).
Figure 4.
The CABS-flex 3.0 web interface presenting example modeling results in Flexibility Modeling mode. The interface displays the “Models” tab, showing a protein model colored according to RMSF.
Project information tab contains details about the job, basic input information, and the options introduced by a user.
Models tab enables users to visualize 3D structural outputs using the mol* viewer [17]. Users can toggle between superimposed view of all 10 models and detailed individual views. This tab offers extensive customization, allowing structures to be colored according to various experimental and simulation-derived metrics such as B-factor, pLDDT, and RMSF, as well as by sequence, chain, or secondary structure attributes. Additionally, users can adjust visual settings, like background, shadows, and outlines. All structural data, including trajectory and output models, are available for download in a single compressed package (.tar.gz).
Contact maps tab integrates interactive Plotly generated contact matrices with a synchronized mol* viewer [17] for an in-depth exploration of residue interactions across the simulation trajectory. It generates contact maps for each combination of chains, including intra-chain interactions, with contact frequencies normalized between 0 and 1 and visually distinguished by color. Hover tooltips provide detailed residue pair information and interaction frequencies, while selecting any contact point dynamically highlights the corresponding residues in the 3D structure and zooms in for closer inspection. This dual-view approach merges sequence-based interaction patterns with structural insights, aiding in the identification of critical inter-residue contacts. Additionally, for analyses related to the physicochemical nature of these contacts, the Mapiya server can be used [18].
Fluctuation plot tab plots RMSF (and pLDDT, if applicable) values per residue as interactive 2D graphs, annotated with secondary structure elements. This detailed visualization enables users to zoom, drag, and hover over data points to access specific information about residue fluctuations. The overlay of secondary structure elements allows for a direct correlation between observed flexibility and distinct structural motifs, making it easier to pinpoint areas of inherent stability as well as regions exhibiting significant conformational variability.
Server architecture and documentation
The CABS-flex 3.0 server features an intuitive Flask-based HTML interface dynamically rendered via Jinja2 templating system. User submissions are validated and stored in a MySQL database, with jobs executed immediately or queued based on resource availability. Users can track computational progress with real-time status updates (pending, queued, running, and completed). Molecular structures are rendered using the Mol* library [17] (HTML5/JavaScript), while interactive contact maps and simulation data are plotted with Plotly (https://plotly.com/). Protein data are fetched directly from the PDB via RESTful APIs. The server is hosted on Apache2 and manages the user queue via MySQL. At its core, CABS-flex 3.0 employs the CABS-flex standalone package [6]. The service is freely accessible without login, with detailed documentation provided on the “How to” page: https://lcbio.pl/cabsflex3/howto.
Summary
CABS-flex 3.0 enables efficient simulations of protein and peptide dynamics using a coarse-grained model combined with all-atom reconstruction. It introduces flexibility modes (Flexible, Rigid, Rigid-pLDDT, and Unleashed), allowing users to control the extent of conformational freedom in simulations. The method integrates pLDDT-derived restraints, improving accuracy by incorporating structural confidence levels. It also supports peptide modeling, enabling structure prediction for both linear and cyclic peptides. Users can input a protein structure or peptide sequence, while output includes simulation trajectories and model ensembles, which can be analyzed via the web interface. Default settings are optimized, but users can adjust flexibility levels based on experimental or computational data. The intuitive web interface, combined with optimized default settings and flexible control options, makes CABS-flex 3.0 a versatile tool for studying protein and peptide dynamics.
Acknowledgements
We would like to express our sincere gratitude to Grzegorz Firlik for his invaluable assistance in setting up the web server infrastructure and for enabling computations on the Funk supercomputer at the Biological and Chemical Research Centre, University of Warsaw. We also extend our thanks to Rafał Madaj for his support during the early stages of web service development..
Author contributions: K.W. (Conceptualization [equal], investigation [equal], methodology [equal], project administration [equal], software [equal], validation [equal], visualization [equal], writing – original draft [equal], writing – review & editing [equal]), M.Z. (Conceptualization [equal], investigation [equal], methodology [equal], software [equal], validation [equal], visualization [equal], writing – original draft [equal]), A.K. (Conceptualization [equal], investigation [equal], methodology [equal], software [equal], validation [equal]), S.K. (Conceptualization [equal], funding acquisition [equal], resources [equal], supervision [equal], validation [equal], writing – original draft [equal], writing – review & editing [equal]).
Contributor Information
Karol Wróblewski, University of Warsaw, Biological and Chemical Research Centre, Faculty of Chemistry, 02-089 Warsaw, Poland.
Mateusz Zalewski, University of Warsaw, Biological and Chemical Research Centre, Faculty of Chemistry, 02-089 Warsaw, Poland.
Aleksander Kuriata, University of Warsaw, Biological and Chemical Research Centre, Faculty of Chemistry, 02-089 Warsaw, Poland.
Sebastian Kmiecik, University of Warsaw, Biological and Chemical Research Centre, Faculty of Chemistry, 02-089 Warsaw, Poland.
Conflict of interest
None declared.
Funding
K.W., M.Z., and S.K. acknowledge funding from the National Science Centre, Poland [2020/39/B/NZ2/01301] OPUS.
Data availability
This website is free and open to all users and there is no login requirement. The web server is available at https://lcbio.pl/cabsflex3. An extended CABS-flex 3.0 description and online tutorials are available at the CABS-flex 3.0 “How to” page: https://lcbio.pl/cabsflex3/howto.
References
- 1. Jamroz M, Orozco M, Kolinski A et al. Consistent view of protein fluctuations from all-atom molecular dynamics and coarse-grained dynamics with knowledge-based force-field. J Chem Theory Comput. 2013; 9:119–25. 10.1021/ct300854w. [DOI] [PubMed] [Google Scholar]
- 2. Jamroz M, Kolinski A, Kmiecik S CABS-flex predictions of protein flexibility compared with NMR ensembles. Bioinformatics. 2014; 30:2150–4. 10.1093/bioinformatics/btu184. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Wróblewski K, Kmiecik S Integrating AlphaFold pLDDT Scores into CABS-flex for enhanced protein flexibility simulations. Comput Struct Biotechnol J. 2024; 23:4350–6. 10.1016/j.csbj.2024.11.047. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Jamroz M, Kolinski A, Kmiecik S CABS-flex: server for fast simulation of protein structure fluctuations. Nucleic Acids Res. 2013; 41:W427–31. 10.1093/nar/gkt332. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Kuriata A, Gierut AM, Oleniecki T et al. CABS-flex 2.0: a web server for fast simulations of flexibility of protein structures. Nucleic Acids Res. 2018; 46:W338–43. 10.1093/nar/gky356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Kurcinski M, Oleniecki T, Ciemny MP et al. CABS-flex standalone: a simulation environment for fast modeling of protein flexibility. Bioinformatics. 2019; 35:694–5. 10.1093/bioinformatics/bty685. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Badaczewska-Dawid A, Wróblewski K, Kurcinski M et al. Structure prediction of linear and cyclic peptides using CABS-flex. Brief Bioinform. 2024; 25:bbae003. 10.1093/bib/bbae003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Nithin C, Fornari RP, Pilla SP et al. Exploring protein functions from structural flexibility using CABS-flex modeling. Protein Sci. 2024; 33:e5090. 10.1002/pro.5090. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Heo L, Feig M One bead per residue can describe all-atom protein structures. Structure. 2024; 32:97–111.e6. 10.1016/j.str.2023.10.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Vander Meersche Y, Cretin G, Gheeraert A et al. ATLAS: protein flexibility description from atomistic molecular dynamics simulations. Nucleic Acids Res. 2024; 52:D384–92. 10.1093/nar/gkad1084. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Webb B, Sali A Comparative protein structure modeling using MODELLER. Curr Protoc Bioinform. 2016; 54:5.6.1–5.6.37. 10.1002/cpbi.3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Jamroz M, Kolinski A Modeling of loops in proteins: a multi-method approach. BMC Struct Biol. 2010; 10:5. 10.1186/1472-6807-10-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Kmiecik S, Jamroz M, Kolinski M Structure prediction of the second extracellular loop in G-protein-coupled receptors. Biophys J. 2014; 106:2408–16. 10.1016/j.bpj.2014.04.022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Blaszczyk M, Kurcinski M, Kouza M et al. Modeling of protein-peptide interactions using the CABS-dock web server for binding site search and flexible docking. Methods. 2016; 93:72–83. 10.1016/j.ymeth.2015.07.004. [DOI] [PubMed] [Google Scholar]
- 15. Ciemny M, Badaczewska-Dawid A, Pikuzinska M et al. Modeling of disordered protein structures using Monte Carlo simulations and knowledge-based statistical force fields. Int J Mol Sci. 2019; 20:606. 10.3390/ijms20030606. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Høie MH, Kiehl EN, Petersen B et al. NetSurfP-3.0: accurate and fast prediction of protein structural features by protein language models and deep learning. Nucleic Acids Res. 2022; 50:W510–5. 10.1093/nar/gkac439. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Sehnal D, Bittrich S, Deshpande M et al. Mol* viewer: modern web app for 3D visualization and analysis of large biomolecular structures. Nucleic Acids Res. 2021; 49:W431–7. 10.1093/nar/gkab314. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Badaczewska-Dawid AE, Nithin C, Wroblewski K et al. MAPIYA contact map server for identification and visualization of molecular interactions in proteins and biological complexes. Nucleic Acids Res. 2022; 50:W474–82. 10.1093/nar/gkac307. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
This website is free and open to all users and there is no login requirement. The web server is available at https://lcbio.pl/cabsflex3. An extended CABS-flex 3.0 description and online tutorials are available at the CABS-flex 3.0 “How to” page: https://lcbio.pl/cabsflex3/howto.